Skip to Main content Skip to Navigation

Analyse et recherche d'oeuvres d'art 2D selon le contenu pictural

Abstract : Cultural institutions around the world have, over the past two decades, implemented a policy of digitally safeguarding their artwork collections, building extremely large image databases. Works of visual art are quite distinct from natural images in that they are often stylized. This property introduces a subjectivity that influences one's understanding of the scene as well as the impression conveyed to the viewer when he/she looks at the image. Such database can benefit from specific content-based features. The first aspect of the pictorial content that we studied concerns the spatial organization of colours within a work of visual art. We proposed a methodology to model the difference between the spatial arrangements of two images as an optimal transportation problem. We also proposed an adaptive matching criterion based on a statistical approach, itself founded on an "a contrario" approach. The second part of my doctoral research focused on characterising the geometrical information and pictorial impact conveyed by linear strokes. The methodology we developed here is based firstly on a novel parameter-less method for the extraction of stroke boundaries in line artworks based on a selection of the level-lines of the topographic map. We also developed a set of methods that analyzes these boundaries to extract several intuitive geometrical cues. We evaluated the performance of these methods on seven different databases comprising a total of about 65000 images.
Document type :
Complete list of metadata

Cited literature [253 references]  Display  Hide  Download
Contributor : Thomas Hurtut Connect in order to contact the contributor
Submitted on : Tuesday, October 27, 2009 - 10:55:07 AM
Last modification on : Friday, January 21, 2022 - 3:29:16 AM
Long-term archiving on: : Thursday, June 17, 2010 - 5:24:13 PM


  • HAL Id : tel-00426645, version 1


Thomas Hurtut. Analyse et recherche d'oeuvres d'art 2D selon le contenu pictural. Interface homme-machine [cs.HC]. Télécom ParisTech, 2008. Français. ⟨tel-00426645⟩



Record views


Files downloads